Skip to contentSkip to conversation

MINSTATE_START_SAVE_MIN_PHYS looking very confused

DISCUSSION David Mosberger 3 participants 5 messages May 6, 2005 – May 10, 2005

lore
Share & tools
lore
#0 lore raw

While working on the libunwind-enabled Linux kernel, I noticed that MINSTATE_START_SAVE_MIN_PHYS is loking rather confused. Part of the problem seems to have been introduced by rja's patch from last December (it cause "r1" to be setup incorrectly when already in kerenl mode). However, the bigger issue to me is: given that we have a separate INIT stack, why bother distinguishing between an INIT that hits user-level vs. one that hits the kernel. It's not possible to handle nested INITs anyhow, so why not just simplify that code to always switch to the INIT stack?

	--david
#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · David Mosberger

David> It's not possible to handle nested INITs anyhow, so why not David> just simplify that code to always switch to the INIT stack?

I suspect Keith will tell me "that's what I've been working on". ;-)

However, the existing code clearly is bogus, putting the kernel stack at a (almost) random place when INIT hits while in kernel-mode. Patch below fixes this.

	--david

[IA64] Fix stack placement when INIT hits in kernel mode.

Without this patch, the stack is placed _below_ the current task structure, which is risky at best.

Tony, I think this patch needs to go into 2.6.12, since it fixes a real bug. Without it, INIT may case secondary errors, which would be most unpleasant.

Index: arch/ia64/kernel/minstate.h
=================================--- b5f77812a2ece5addfd983e38eda85017039a29d/arch/ia64/kernel/minstate.h (mode:100644 sha1:1dbc7b2497c90d4b3c0a8cef4c7afb94e7abc315) +++ uncommitted/arch/ia64/kernel/minstate.h (mode:100644)
@@ -41,7 +41,7 @@
(pKStk) addl r3=THIS_CPU(ia64_mca_data),r3;; \
(pKStk) ld8 r3 = [r3];; \
(pKStk) addl r3=IA64_MCA_CPU_INIT_STACK_OFFSET,r3;; \
-(pKStk) addl sp=IA64_STK_OFFSET-IA64_PT_REGS_SIZE,r3; \
+(pKStk) addl r1=IA64_STK_OFFSET-IA64_PT_REGS_SIZE,r3; \
(pUStk) mov ar.rsc=0; /* set enforced lazy mode, pl 0, little-endian, loadrs=0 */ \
(pUStk) addl r22=IA64_RBS_OFFSET,r1; /* compute base of register backing store */ \
;; \
@@ -50,7 +50,6 @@
(pUStk) mov r23=ar.bspstore; /* save ar.bspstore */ \
(pUStk) dep r22=-1,r22,61,3; /* compute kernel virtual addr of RBS */ \
;; \
-(pKStk) addl r1=-IA64_PT_REGS_SIZE,r1; /* if in kernel mode, use sp (r12) */ \
(pUStk) mov ar.bspstore=r22; /* switch to kernel RBS */ \
;; \
(pUStk) mov r18=ar.bsp; \

#2 lore raw
In reply to David Mosberger
Quoted 6 lines · David Mosberger

However, the bigger issue to me is: given that we have a
separate INIT stack, why bother distinguishing between an INIT that hits user-level vs. one that hits the kernel. It's not possible to handle nested INITs anyhow, so why not just simplify that code to always switch to the INIT stack?

I don't know the historical reason for the user/kernel differences in the code. Maybe concerns about saving the right state information depending on the context? You're right that it could be made to treat both the same.

Signature · 2 lines
Russ Anderson, OS RAS/Partitioning Project Lead  
SGI - Silicon Graphics Inc          [email protected]
#3 lore raw
In reply to David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Russ Anderson
Quoted 5 lines · Russ> David Mosberger wrote:
Quoted 5 lines

However, the bigger issue to me is: given that we have a
separate INIT stack, why bother distinguishing between an INIT that hits user-level vs. one that hits the kernel. It's not possible to handle nested INITs anyhow, so why not just simplify that code to always switch to the INIT stack?

Russ> I don't know the historical reason for the user/kernel Russ> differences in the code. Maybe concerns about saving the Russ> right state information depending on the context?

I don't know the history there either. At least in today's environment with per-CPU INIT stacks, the code makes little sense, though.

Keith, are your MCA cleanups close to being ready for prime-time?

	--david
#4 lore raw
In reply to David Mosberger
Quoted 1 line · David Mosberger

Keith, are your MCA cleanups close to being ready for prime-time?

Stalled for four weeks, due to SLES9 work and the changeover from bk to git. I hope to get back to the MCA changes this week.